Create your first agent
Define a support agent in Agent Builder that is narrow enough to stay safe and useful enough to carry the first wave of live traffic.
Start with one clear job
The first agent should own a tight support scope: order status, shipping expectations, FAQ answers, and common policy guidance. Avoid broad promises about every workflow in the business.
A smaller role makes it easier to review transcript quality and see where the agent needs more knowledge or a safer escalation rule.
Set persona and escalation early
- 1In Setup (Persona & Identity), write the agent as an operator, not a marketer. Keep answers direct and grounded.
- 2Add explicit escalation moments for disputes, account risk, or answers the agent cannot support with evidence.
- 3Use a clear name so the team recognizes which agent owns incoming traffic.
Good first-pass behavior
Helpful, concise, evidence-based, and willing to escalate instead of improvising.
Do the minimum ready check
- The name and purpose are obvious to the team.
- The persona behavior matches support, not generic AI chat.
- Escalation rules are visible and understood before you connect a channel.
Should I create the agent before I connect a channel?
Yes. Build the agent first so the first live traffic lands on a workspace persona with clear tone, escalation rules, and knowledge attached.
What is the safest way to decide when the agent should escalate?
Start narrow. Escalate on refund disputes, sentiment spikes, account risk, or any place where a wrong answer creates operational cost. Expand only after reviewing real transcripts.
Launch your workspace
The operator path from a brand-new workspace to the first healthy live conversations, working through Agent Builder in order without skipping the validation that protects answer quality.
Connect your first channel
Use the Connections section to bring in one live support channel, verify health, and confirm the route lands on the intended workspace without hidden gaps.
Add website and FAQ knowledge
Use the Knowledge section to ground the agent in one reliable help source so answers stay specific instead of slipping into generic support language.